Reminding about Tagged Objects using Passive RFIDs

1. Reminding about Tagged Objects using Passive RFIDs Gaetano Borriello, Waylon Brunette, Matthew Hall, Carl Hartung, and Cameron Tangney

2. Requirements for the application • Ability to tag many items cost-effectively • Tags should need only minimal maintenance • Number and intrusiveness of reminders should be minimized • User’s current location and destination provides useful context • Easy to add reminder rules • Incrementally deployable and easy to maintain 2004 UW CSE Affiliates Meeting

3. Our work (outline) • Are passive RFIDs appropriate for this application? • Passive RFID tags on objects – no batteries, small • Can infrastructure requirements of RFIDs be acceptable? • Broadcasting RFID readers not connected to network • What is an appropriate UI for the user? • Wristwatch user interface for alarms and reminders • Auto-tags for most common rules/behaviors • What other context is needed for good reminders? • Locations from readers and user’s calendar data 2004 UW CSE Affiliates Meeting

4. RFID tags • Alien ALL-9250 915MHz long-range RFID • 15.5 x 1.6 x .03 cm • 3-4m read range • 64 bits 2004 UW CSE Affiliates Meeting

5. RFID readers • Alien readers • Up to 4 antennas per reader (mux’ed) • Serial interface reports reads • Can set read rate • UCB Mica sensor mote • Broadcasts tag read data • 433MHz 38.4Kbps • 15m range • Also broadcast reader’s location • No connection to infrastructure + 2004 UW CSE Affiliates Meeting

6. What the user carries (besides tagged objects) • Personal server • Intel Research • XScale/Linux • UCB mote RF interface • Runs application • Wrist-watch • 64x128 B/W LCD • Atmel microcontroller • 5 buttons for input • UCB mote RF interface • Provides user interface • Screen and beeper 2004 UW CSE Affiliates Meeting

7. user passes by a reader antenna on wall alarm if object that should be present is not System overview tags on objectsand server on wrist 2004 UW CSE Affiliates Meeting

8. Experiments (system) • single-user scenario • 5 locations • 8 objects • user’s calendar data keys, wallet, phone,jacket, backpack, keycard, papers, gym bag 2004 UW CSE Affiliates Meeting

9. Results for tag experiments • ~100% success rate for non-metal objects • coats, papers, gym bag, etc. • 50-80% for metal objects • keys, laptop, cell phone, etc. • Solution: 1-2mm standoff get this to almost 100% • 25-40% when tags in contact with or very near skin • objects in front jeans pocket or in hand, e.g., cell phone • Solution: manufacture object with internal tag 2004 UW CSE Affiliates Meeting

10. 25 20 15 10 5 0 0% 5% 10% 15% 20% 25% 30% 35% 40% Robustness simulations – varying read misses randomly # of alarms miss rate 2004 UW CSE Affiliates Meeting

11. Auto-tag • Directly implement common situations • Explicitly adds rule to database • No user programming required • Auto-tag as a Post-It note • Auto-tags are registered to user/location pair so that they are recognized by application (not confused with other users’ objects) • Add to object • Indicates to application it must be returned to registered location of the tag 2004 UW CSE Affiliates Meeting

12. Privacy issues • Readers do not store tag reads • No connection to infrastructure • Nearby user’s tags are also read and broadcast • Ignored by other users • No connection to owner’s object data base • Tag IDs can be scrambled • daily (hourly) by user’s own reader(s) – e.g., home and car 2004 UW CSE Affiliates Meeting

13. Conclusion • RFID tags are a good solution for object reminding • Readers only need to broadcast their tag reads into local space– no centralized infrastructure necessary • Incrementally deployable • Less vulnerable to privacy problems (data not stored) • Wrist-watches are interesting user interfaces for ubiquitous computing systems • Future work: integrating location and inference 2004 UW CSE Affiliates Meeting